# OCR Medical Records for Attorneys: A Working Method

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2023-10-14T03:18:05Z
- Updated: 2026-10-11T19:22:20Z
- Category: Practice areas
- Canonical: https://www.schemalaw.com/blog/ocr-medical-records-for-attorneys

> A working method to OCR medical records for attorneys: split the set by provider, convert it, check what matters, and build a chronology tied to pages.

To OCR medical records, attorneys should keep the set exactly as the provider sent it, split a working copy by provider into files small enough to upload, run recognition, and then check dates, names and dosages against the page images. The result is searchable text you can build a chronology from, with every entry tied to a page.

The method below assumes a typical personal injury or disability file: several hundred pages from a handful of providers, some of it faxed, some of it printed from an electronic chart. It works the same way for a workers' compensation claim or a malpractice review. Only the volume changes.

Incident reports often sit under odd headings. Search the converted set for patient safety notes, then confirm the dates on the image. The same chart may be used in a personal injury case, a disability claim, or a workers compensation proceeding. Label the working copy so a duty of confidentiality review covers the same pages as the original chart.

## Why are medical records hard to work with as scans?

Medical records are hard to work with as scans because they are long, repetitive, out of order and built from many page designs at once. A records custodian prints or faxes whatever the chart holds, and the firm receives one image-only PDF in which nothing can be searched and little follows the calendar.

Four features cause most of the trouble.

- **Volume with little signal.** A printout from an electronic chart repeats the medication list, the allergy list and the problem list on visit after visit. The two sentences that describe how the patient was actually doing sit in the middle of a page of boilerplate.
- **No reliable order.** Some custodians sort by document type, so every lab result comes before every office note. Others send the newest visit first. A hospital stay may arrive as nursing flowsheets, then physician orders, then the discharge summary that explains them.
- **Mixed layouts.** One set can hold typed narrative, forms with checkboxes, lab tables, medication grids, fax cover sheets and a handwritten intake questionnaire. Each of those asks something different of recognition software.
- **Copies of copies.** Records often reach you after a fax, a rescan and a print. Small type in a lab table suffers most. The checklist for [low-quality PDF scans](https://www.schemalaw.com/blog/ocr-low-quality-pdf) helps you decide which pages are worth asking the provider to send again.

There is also a vocabulary problem. Charts are written in abbreviations, and many are two or three letters long. A general spell checker has no opinion on whether "bid" should have been "tid", so a recognition error in a dosing instruction looks like a perfectly good word.

## What does OCR make possible with medical records?

OCR makes it possible to search a records set, copy passages out of it and sort its contents by date, none of which an image-only scan allows. For medical records that turns a page-by-page read into a targeted one: you can find every mention of a body part, a drug or a provider in seconds.

In practice the gains fall into four groups.

1. **Search across the whole set.** Look for "lumbar", "radiculopathy" or the name of the treating surgeon and jump to each hit. Search for the weeks before the incident to find prior complaints, which is usually the first thing the other side will look for.
2. **Text you can quote.** A sentence from an operative report can be pasted into a demand letter or a deposition outline. Read it against the page before you rely on it.
3. **Tables as tables.** Lab panels and medication lists can come out as rows and columns instead of a run of numbers. Billing ledgers are the most valuable case, and the guide to [converting medical records to a spreadsheet](https://www.schemalaw.com/blog/how-to-convert-medical-records-to-a-spreadsheet) covers them.
4. **A page reference for everything.** Converted text keeps its place in the file, so a chronology entry can point to the page it came from.

OCR does not interpret the chart. It will not tell you that a finding is significant, reconcile two providers who disagree, or notice that a visit is missing. It gives a reviewer faster access to what the pages say. The medical judgment still belongs to the person reading.

## A method for converting a records set

Convert a records set in six steps: preserve the original, split a copy by provider, name the pieces, upload them, check the output, and export in the formats the next task needs. The order matters, because every later citation depends on page numbers that do not move.

1. **Preserve what you received.** Save the provider's file untouched, with the date it arrived and how it was requested. Work only on copies. If the pages carry Bates numbers, or you plan to add them, decide that now. The post on [how to OCR Bates numbered documents](https://www.schemalaw.com/blog/how-to-ocr-bates-numbered-documents) explains the sequence.
2. **Split by provider, then by size.** One file per provider is the natural unit, since each provider's pages share a layout. Schema Law accepts PDF, PNG or JPEG files up to 20 MB. The Solo plan takes up to 250 pages per document and the Team plan up to 1,000, so a long hospital chart may need to be cut into parts. Cut at a document boundary, never in the middle of a report.
3. **Name each part so the page math is recoverable.** Exports number pages from 1 within each uploaded file. If a hospital chart becomes three parts, put the original page range in the file name, such as `StMarys_part2_p251-500.pdf`. Page 14 of that export is page 264 of the set.
4. **Upload and wait for the pages to be read.** Sign in, add the files, and let each one finish. The [three-step overview on the homepage](https://www.schemalaw.com/#how-it-works) shows the sequence, and says there is nothing to prepare before you upload.
5. **Check the output against the images.** Sample every provider, and read every page you intend to cite. The next two sections say where to look hardest.
6. **Export for the task.** A searchable PDF is the copy to keep in the file and hand to an expert. Word or plain text suits quoting. Excel suits anything with columns.

### A page log that saves you later

A short log, kept as you split, answers the question "where did this page come from" for the life of the matter.

| Working file | Provider | Pages in the set | Received | Notes |
| --- | --- | --- | --- | --- |
| `StMarys_part1_p001-250.pdf` | St. Mary's Hospital | 1 to 250 | 3 March | Emergency department and admission |
| `StMarys_part2_p251-500.pdf` | St. Mary's Hospital | 251 to 500 | 3 March | Nursing flowsheets, mostly tables |
| `Ortho_p501-612.pdf` | Valley Orthopedics | 501 to 612 | 11 March | Faxed; pages 540 to 548 faint |
| `PT_p613-701.pdf` | Ridge Physical Therapy | 613 to 701 | 19 March | Handwritten visit notes |

The names and dates are invented. The columns are the point: with them, any export page converts back to a page in the set as produced.

On volume, the 701 pages in this example fit within the 1,000 pages a month of the Solo plan. A firm that handles sets this size every week should read the [pricing page](https://www.schemalaw.com/pricing) first, and the post on [batch OCR for discovery documents](https://www.schemalaw.com/blog/batch-ocr-discovery-documents) for how to keep many parts in order.

## Handwritten chart notes and provider signatures

Handwritten chart notes are the least dependable part of any converted record, and a signature should be treated as a mark, not as text. Expect recognition to do well on typed narrative and badly on a hurried progress note. Plan to read handwritten pages yourself and to transcribe the passages you need.

Handwriting turns up in predictable places: intake questionnaires the patient filled in, pain diagrams, physical therapy visit notes, older office charts, consent forms, and corrections written over a typed line. Mark those pages in your log as you find them. The post on [what OCR can do with handwritten notes](https://www.schemalaw.com/blog/ocr-handwritten-notes) explains why they behave differently and how to get the best from them. Test your own pages before you assume anything about a particular hand.

Three habits keep handwritten material from causing an error in your work product.

- **Never quote recognized handwriting without reading the image.** A plausible wrong word is worse than a blank.
- **Record the gap.** If a word cannot be made out by a person either, write "[illegible]" in your chronology. Do not guess.
- **Ask the source.** A treating provider can often supply a typed version of a note, or dictate what it says. For an important entry, that is better evidence than anyone's reading of a scrawl.

Signatures need a different kind of care. Many electronic charts print a typed line such as "Electronically signed by" followed by a name and a time, and that line converts like any other text. A pen signature is a drawing. Software may skip it or turn it into a few stray letters. If the identity of the signer matters, take it from the typed name block beneath the signature or from the page image, and say which.

Checkboxes and circled answers deserve the same suspicion. A tick beside "Yes" on a prior-injury question is easy for a person to see and easy for software to misplace. Confirm every form answer you rely on from the image.

## Building a medical chronology from converted records

Build a medical chronology by pulling one row per encounter from the converted text, sorting the rows by date of service, and keeping a page reference on every row. Search finds the encounters. A person decides what each one means and writes the summary.

A chronology that holds up usually has these columns.

| Date of service | Provider | Type of record | What it says | Source |
| --- | --- | --- | --- | --- |
| 14 Jan | St. Mary's Hospital | Emergency department note | Complaint, findings, treatment, discharge instructions | Set p. 12 to 15 |
| 22 Jan | Valley Orthopedics | New patient visit | History given, exam, imaging ordered, work status | Set p. 503 to 506 |
| 2 Feb | Valley Imaging | MRI report | Impression section quoted in full | Set p. 588 |

A workable sequence:

1. **Find the encounters.** Search each converted file for phrases that begin a visit in that provider's layout, such as "Date of Service", "Encounter Date" or "Admit Date". The phrase differs by provider, so look at three or four pages first.
2. **Log each one.** Enter the date, the provider, the type of record and the pages. Do this for the whole set before you summarize anything, so that gaps in treatment show up early.
3. **Summarize from the image.** With the page open, write what the note says about complaints, findings, diagnosis, treatment and restrictions. Quote the provider's words where they matter.
4. **Sort and review.** Order by date of service and read down the column. A date far out of sequence is usually a typing or recognition error.

Dates need attention of their own. One page can carry a date of service, a date the note was signed, a date it was printed and a fax transmission date along the top edge. Only the first belongs in the date column. A report dictated a week after a visit should be listed under the visit, with the dictation date in the summary if it matters.

Repetition is the other trap. A history copied forward from visit to visit can make one complaint look like ten. Enter the history where it first appears and note later visits only for what changed.

Check numbers character by character: doses, vital signs, lab values, and left against right. A decimal point lost in a fax changes a dose tenfold. The post on [why OCR output is full of errors](https://www.schemalaw.com/blog/why-is-ocr-output-full-of-errors) describes a proofreading method that fits this kind of review.

## Confidentiality when records leave the firm

Medical records are among the most sensitive papers a firm holds, so decide whether they may go to any online service before you upload one page. That decision rests on your own obligations: professional rules, the terms on which you received the records, any protective order, and privacy law where it applies to you.

Schema Law makes no claim to comply with HIPAA or any other health privacy standard. Do not read one into this post. Whether a set of records may be uploaded to Schema Law, or to any other online tool, is a question for you to answer under the duties that bind your firm.

One point of federal law is worth knowing as you think it through. The HIPAA regulations define a ["business associate" in 45 CFR 160.103](https://www.law.cornell.edu/cfr/text/45/160.103) to include a person who provides legal services to a covered entity where that work involves protected health information, and they extend the term to a subcontractor that receives such information on the business associate's behalf. A firm that acts for a hospital or a health plan is in a different position from one that holds a client's own records under a signed authorization. Work out which you are before you choose a tool. This post does not give legal advice.

Questions to settle first:

- Under what authority did the records come to you, and does it limit who may handle them?
- Does a protective order or confidentiality agreement cover the set?
- Do the records include categories that often carry extra protection, such as mental health or substance use treatment? Rules for those vary by jurisdiction.
- What does the vendor say in writing about training, retention and deletion?
- Can you remove the files when the work is done?

What Schema Law does state appears in the [lines it lists for every plan](https://www.schemalaw.com/#security): documents are never used to train an AI model, they stay inside your own account, they are encrypted in transit and at rest, and you can delete any document at any time. Read them as written and weigh them against your own requirements.

For the wider analysis, see the posts on [attorney-client privilege and AI tools](https://www.schemalaw.com/blog/attorney-client-privilege-and-ai-tools) and on [whether OCR software stores your documents](https://www.schemalaw.com/blog/does-ocr-software-store-my-documents). If the answer for a given set is no, a desktop program that runs on the firm's own machines is the alternative.

## Try it on a real document

Upload a scanned PDF and export it as Word, a searchable PDF or plain text on the free plan. Your documents are never used to train an AI model. [Start free](https://www.schemalaw.com/api/auth/google)